Re: [Qemu-devel] Why qemu write/rw speed is so low?

[Zhi Yong Wu]

On Tue, Sep 13, 2011 at 4:49 PM, Stefan Hajnoczi <address@hidden> wrote:
> On Tue, Sep 13, 2011 at 9:31 AM, Zhi Yong Wu <address@hidden> wrote:
>> On Tue, Sep 13, 2011 at 3:15 PM, Stefan Hajnoczi
>> <address@hidden> wrote:
>>> On Tue, Sep 13, 2011 at 10:38:28AM +0800, Zhi Yong Wu wrote:
>>>> On Fri, Sep 9, 2011 at 6:38 PM, Stefan Hajnoczi
>>>> <address@hidden> wrote:
>>>> > On Fri, Sep 09, 2011 at 05:44:36PM +0800, Zhi Yong Wu wrote:
>>>> >> Today, i did some basical I/O testing, and suddenly found that qemu 
>>>> >> write and rw speed is so low now, my qemu binary is built on commit 
>>>> >> 344eecf6995f4a0ad1d887cec922f6806f91a3f8.
>>>> >>
>>>> >> Do qemu have regression?
>>>> >>
>>>> >> The testing data is shown as below:
>>>> >>
>>>> >> 1.) write
>>>> >>
>>>> >> test: (g=0): rw=write, bs=512-512/512-512, ioengine=libaio, iodepth=1
>>>> >
>>>> > Please post your QEMU command-line.  If your -drive is using
>>>> > cache=writethrough then small writes are slow because they require the
>>>> > physical disk to write and then synchronize its write cache.  Typically
>>>> > cache=none is a good setting to use for local disks.
>>>> >
>>>> > The block size of 512 bytes is too small.  Ext4 uses a 4 KB block size,
>>>> > so I think a 512 byte write from the guest could cause a 4 KB
>>>> > read-modify-write operation on the host filesystem.
>>>> >
>>>> > You can check this by running btrace(8) on the host during the
>>>> > benchmark.  The blktrace output and the summary statistics will show
>>>> > what I/O pattern the host is issuing.
>>>>   8,2    0        1     0.000000000   337  A  WS 425081504 + 8 <-
>>>> (253,1) 42611360
>>>
>>> 8 blocks = 8 * 512 bytes = 4 KB
>> How do you know each block size is 512 bytes?
>
> The blkparse format specifier for blocks is 'n'.  Here is the code to
> print it from blkparse_fmt.c:
>
> case 'n':
>    fprintf(ofp, strcat(format, "u"), t_sec(t));
>
> And t_sec() is:
>
> #define t_sec(t)        ((t)->bytes >> 9)
Great, it shift 9 bit in the right direction, i.e. its unit is changed
from bytes to blocks, got it, thanks.

>
> So it divides the byte count by 512.  Block size == sector size == 512 bytes.
>
> You can get the blktrace source code here:
>
> http://brick.kernel.dk/snaps/
>
> Stefan
>



-- 
Regards,

Zhi Yong Wu